OBJECTIVE This study aimed to critically investigate all-cause and major-cause mortality of incident rheumatoid arthritis (RA) cases versus matched non-RA comparison (CN) subjects in a long-term prospective cohort. METHODS Baseline 1974 cohort entry demographic and serum biomarker data on 54 incident RA patients and 216 matched CN subjects were related to their mortality from 1995 through 2015. Mortality of RA patients was also analysed by 3 categories of course responses to therapy assigned by the sole community rheumatologist in 1995 (19 good, 23 fair, and 12 limited). Cox proportional hazards regression models including baseline covariates were used to determine survival from all-causes, cardiovascular disease (CVD), respiratory-related, malignancies, and other causes of death (CODs). RESULTS Total deaths occurred in 38 (70.4 percent) of 54 RA and 102 (47.7 percent) of 216 CN (p=0.003). Total mortality remained greater (p=0.011) in RA versus CN subjects after adjustment for baseline demographic covariates (HR= 1.66, 95% CI 1.12-2.46). Respiratory-related CODs were also greater (p=0.047) in RA versus CN (HR= 2.69, 95% CI 1.02-7.14) subjects. The RA patients' responses to therapy in 1995 significantly (p=0.004) predicted total mortality. Baseline serum immunological and steroid biomarkers independently predicted total, CVD, and other and unknown CODs. Pre-clinical (1974) ranked biomarker z-score values (1 = lowest, 5 = highest) within matched sets of 1 RA and 4 CN study subjects independently associated with mortality from 1995 through 2015, for both total (CRP, p=0.028 and sIL-2Rα, p=0.030) and CVD (CRP, p=0.005 and sTNF-R1, p=0.003) deaths. CONCLUSIONS Total mortality and respiratory-related CODs were greater in incident RA versus CN subjects. The 35 RA cases who had fair or limited course responses to rheumatologist's therapy had greater mortality than their matched CN, whereas the 19 good RA responders had equivalent survival to CN subjects. The independent CRP and sTNF-R1 biomarker associations with CVD deaths were enhanced by a gradient of their dichotomous z-score values in survival models.